The application of uniform conservation schemes often fails to account for small-scale spatial variation in the drivers of population decline. Demographic comparisons of imperilled populations across locations are therefore crucial for successful conservation, but progress is hampered by lack of long-term data from more than a single population. The recent large-scale decline of eider ducks (Somateria mollissima) in the Baltic Sea is ideal for determining to what extent mechanisms underlying population decline can be extrapolated over larger areas. We utilized stochastic demographic methods incorporating both environmental and sampling variation to assess small-scale spatial and temporal variation in the population dynamics of eiders at Söderskär (eastern range-margin) and Tvärminne (core breeding area), situated 130 km apart. The stochastic growth rate models accurately predicted the observed differences in the rate of decline between sites and time periods. At Söderskär, established breeder survival had by far the greatest elasticity, whereas elasticity was more evenly distributed among vital rates at Tvärminne. Although the study sites showed the single largest difference in fecundity, stochastic life table response experiment analyses revealed that reduced adult female survival at Tvärminne mainly determined the observed difference in growth rates between sites. In contrast, reduced fecundity primarily differentiated the past population increase from the present population decline at Söderskär. Our results demonstrate that different mechanisms may be associated with population decline across adjacent geographic locations, and indicate that dispersal of first-time breeders may be important for population dynamics. Safeguarding adult female survival and/or fecundity should be prioritized in management efforts.